219
220

This week Ai has killed one more thing, my passion to mentor internsAI/LLM (self.ExperiencedDevs)

submitted by kr0n0sShrugg3d

Yet another rant on Ai.

Reposting because Ai related content is allowed on Saturdays, so last post got deleted.

I am a senior developer with 8 years of experience at one of tier 0 services at the biggest cloud provider ( cannot be more specific than that :/). And yes this post is not written by Ai because that would defeat the entire purpose.

Over the years one of my most passionate hobbies was mentoring college interns and helping them onboard to the ✨amazing,✨ world of software development and because having an offer from such a huge company at beginning of your career can be life changing. Code reviews for interns have been an all time favourites of mine, it is very refreshing to sit next to someone who has not written production grade code, understand their thinking behind it, explaining how it can be better and what design patterns to use and untangle confusion about IAC. More often then not interns would make me question my fundamentals, clarify them and make me better.

But something changed this year, Ai got competent.

The interns are not writing any code by themselves or exploring repositories, they are not able to explain why something is written in the way it is. There are no more discussions on comments or learnings from them. When I ask them a question they put it on a chatbot and paste what the chatbot says. When I tell them not use Ai at the first pass and learn by themselves they would keep on doing it anyway. Debugging is comical to watch as they are not able to navigate through a stack trace and just use Ai for that. They take what Ai says at face value and just spit it out. (Reminds me of that meme where a child is yelling slurs , but it is their parents tongue talking through them)

They are missing the point that entire internship is for them to learn. If I wanted to get something done via Ai I would have done it myself. If your Code reviews is just code by Ai, if your answers are just answers by Ai then why don't I cut out the middle man and talk to Ai?

Don't get me wrong I am not anti Ai, I have been using it to 10x my productivity and offload testing and repetitive tasks and code as well as per my specifications. I am just scared for the next cohort of devs.

Perhaps my fears are unfounded, Ai will keep on improving exponentially, there is a lot of money riding on it. Perhaps this is just the dead canary indicating a bigger gas leak that will silently kill this profession.

But for now I mourn the death of this passion of mine.

all 130 comments
sorted by:
best
topnewcontroversialoldq&a
formatting helphide helpcontent policy

[–]PixelPhoenixForce 161 points162 points  (29 children)

in my team juniors dont even want to mentored anymore - they say AI does better job

[–]kr0n0sShrugg3d[S] 65 points66 points  (22 children)

Ai does explain things really well, you can ask it stupid questions without any fear, but the issue arises when you expect it to do your job without understanding why and what it is doing.

[–]jakubkoneckiSoftware Architect | 25 YE 49 points50 points  (1 child)

I always thought the role of a mentor is to show the person what new questions they can ask - the famous "unknown unknowns".

If you don't know about something, you can't ask AI to explain it to you.

A good mentor will widen the horizon and open doors to areas you haven't even considered before.

[–]kr0n0sShrugg3d[S] 13 points14 points  (0 children)

Well put,that is a good point, and a learning for me.

You will not know what you are seeing unless someone tells you what to look for.

[–]chickadee-guy 20 points21 points  (9 children)

AI does explain things really well, you can ask it stupid questions without any fear

What?? No you cant? It will confidently lie and tell you your shit is golden

[–]kr0n0sShrugg3d[S] 6 points7 points  (2 children)

Even with a human, you always need to verify. But In a lot of cases with Ai you can get working examples you can play around with.

[–]chickadee-guy -5 points-4 points  (1 child)

But In a lot of cases with Ai you can get working examples you can play around with.

Not following at all.

It has a much higher chance to give you a completely incorrect example, or one that is inefficicent, overly verbose and filled with emojis, or from an outdated version of the documentation. And if you dont know how to read code, youre just gonna copy and paste it.

[–]kr0n0sShrugg3d[S] 2 points3 points  (0 children)

I have faced this particular issue with ChatGPT, but not Claude, especially with the 4.x models and proper RAG. Without any context it will surely make up stuff but if you ask it to search the web, compile resources and answer based on that it has been quite helpful.

One good way of testing this out is code exploration when using a new library, but yes I have to sometimes I sister on not making up any details. I do not take its word as the gospel, but as a jumping off pad.

[–]mothzilla 11 points12 points  (3 children)

Maybe they meant asking questions without fear of political blowback from people who think you should already know the answer.

[–]chickadee-guy 6 points7 points  (1 child)

Wanna know whats even worse? Telling those same people to review your AI slop in a PR!!

[–]mothzilla 1 point2 points  (0 children)

Tell me about it.

[–]DigmonsDrill 0 points1 point  (0 children)

If anything the AI will do the opposite and glaze you about what a smart question you had.

Some day the AI will say "wtf, how do you not know this already???" and then the replacement of tech workers will be complete.

[–]anubus72 1 point2 points  (0 children)

how does this shit get upvoted. It depends on how you ask the AI. They are great at code reviews now

[–]dontquestionmyactionSoftware Engineer 0 points1 point  (0 children)

They've gotten a lot better at not doing that so much anymore.

[–]hangfromthisone 0 points1 point  (0 children)

You should then be mentoring on how to use AI properly. Explain that it needs context and it cannot guess or read your mind

[–]gfivksiausuwjtjtnv 8 points9 points  (0 children)

That’s when you reject their PRs until it’s up to the correct standard.

[–]Cultural_Fan_6006 0 points1 point  (0 children)

what's the context for this?

[–]Fearless_Weather_206 0 points1 point  (0 children)

More like AI won’t snitch on you for asking for help every step of the way. you just look like your programming the whole time if no one is looking. Imposter syndrome for reals . Question is are they really absorbing the details or just prompting from one thing to the next? Probably 90% will cheat and pretend while 10% are actually learning from AI and absorbing. Base in the idea that 10% of the company’s employees does the majority of the lifting, while the rest rides off their coattails

[–]big-papito 0 points1 point  (0 children)

There are two skills AI can't teach - debugging, and reading other people's code.

[–]Business_Average1303 -1 points0 points  (0 children)

it does 

[–]PressureAppropriate -3 points-2 points  (0 children)

But it’s true…they’ll be unemployable in a few months and they know it… and so are you by the way…

[–]coordinationlag 31 points32 points  (0 children)

The interns aren't broken, they're just following the actual incentive gradient. The stated goal is "learn" but the real evaluation signal is "did you deliver working code on time." So of course they optimize for delivery speed. AI is the path of least resistance and honestly it's the rational choice if output is what gets you a return offer.

I've seen this exact dynamic before AI even existed. Teams that say they value quality but measure velocity always end up with people gaming the metric. OKRs, sprint points, lines of code, doesn't matter. People optimize for whatever actually gets rewarded.

The thing is, telling them to stop won't work. We tried that at a previous company with a different problem (copy-pasting from Stack Overflow without understanding it) and it just went underground. What actually moved the needle was changing the eval. Make them explain tradeoffs live, walk through a bug they hit, articulate why one approach beats another. That's hard to outsource to a chatbot.

[–]zambizzi 33 points34 points  (0 children)

Man, the angst out there is hitting a fever pitch. I’ve had my fair share as well, and am trying to navigate the changing landscape.

I’m close to 30 years in and I’m an Enterprise Architect at a midsize fintech outfit. I manage a team and still produce quite a bit of code, despite a large pile of responsibilities around that.

We’re being forced into agents writing most or all of the code.

I get the economics of it; it’s faster and it’s “good enough”. If you worked at a car manufacturing plant today, you might tend to robotics that do most of the work. 40 years ago there were more people on the floor, doing this work by hand. Not a perfect analogy due to a factory floor being deterministic, I know, but the point is - you wouldn’t walk onto that job today and start assembling cars by hand. You’d get fired as it’s not your job.

So, for me, I continue to write code by hand outside of work, in addition to continuing learning across all aspects of the job, from system design, architecture, patterns, frameworks, etc. to soft skills and the human side.

You will always need ALL of this to be at your best in this profession! Those who don’t exercise the discipline will not remain competitive. If you’re a junior or student right now and you’re not truly passionate, you’re going to wash out and move on. And, that’s okay. “Everyone learn to code!” was always asinine career guidance.

I’m also not content to hand my career and brain over to Big Tech and these massively overpriced frontier models. I’m spending a lot of time messing with open source models and tools that aren’t quite as good, but they will be, and with decent hardware, don’t cost you a thing. This is the future of the tech IMO, and I think we’ll have huge, empty data centers in a few years, and insanely cheap, powerful hardware at the consumer level.

I’ve mourned and I’ve accepted what’s happening at the professional level. Embrace the suck, adapt, stay sharp and competitive, then build something that allows me to exit the industry and retire.

[–]aruisdanteSoftware Engineer (14 YOE) 70 points71 points  (14 children)

AI is turning everyone into engineering mangers, rather than engineers. Learning how to decompose and delegate a problem, then be “the decider” when the people (agents) implementing it come back with competing solutions and generally guiding evolution of a product forwards without actually implementing it are the skills this cohort values.

So this is what you’re seeing. You’re getting the experience of what a “junior leadership” camp counselor or whatever experiences, instead of the experience of teaching a new generation of engineers.

 They are missing the point that entire internship is for them to learn. If I wanted to get something done via Ai I would have done it myself. If your Code reviews is just code by Ai, if your answers are just answers by Ai then why don't I cut out the middle man and talk to Ai?

While morally I’m on your side as an old timer, I will say that perhaps you’re missing the point from their perspective. To them, the point of this internship is to have you teach them how to use AI better. It isn’t to learn how to write code at all, because as you so rightly point out, in their mind, soon nobody will actually write code themselves any more. They’re betting on a future where software companies will gradually be essentially entirely product people and “management”, and all actual implementation will be done by AI, because AI is the ultimate “low cost off shoring” provider.

They have no interest in learning how to write production code themselves, because they don’t think that that’s actually a valuable skill for them to learn.

[–]kr0n0sShrugg3d[S] 40 points41 points  (6 children)

That is an interesting perspective and it surely does feel that way, but the best EMs I have worked with are those who had been engineers for long times and are very technical. Perhaps we need to find new ways to get that hands-on knowledge?

[–]aruisdanteSoftware Engineer (14 YOE) 13 points14 points  (2 children)

I totally agree. This is the part that’s missed in a lot of the “AI is eating the world” discourse. If you just gave an L3 a team of 12 L3’s under them, it doesn’t magically make them an L6. Only experience actually doing the thing does that, and there’s no way to fast forward that experience. Instead, you would get the exact slop that many engineers turn out blindly trusting AI as it is forced upon them by upper management. The engineers effectively using AI are generally the ones that already had the skills and experience to lead a team.

It’s actually really fascinating to watch the “thought leadership” discourse on AI gradually shift more and more towards essentially being the same advice they give managers in manager training.

And indeed, a lot of the problems companies are facing with AI scaling are the same problems that companies have historically faced when they’ve scaled human workforces too quickly. It’ll be really interesting to see if any research comes out comparing an “all in on AI” company of 100 people to a “legacy” company of 10,000 people. I would bet there is a high degree of overlap in the types of problems they experience.

[–]Akthrawn17 9 points10 points  (1 child)

I will point out that your example of the L3's without AI the slop they created is their slop. They understand it. It might not be optimal, or effective, but it is theirs. And if it fails they can learn from it. THAT is how we "level up" in this industry.

It isn't our successes that provide experience, it is our failures. LLMs are removing the step of failing, thus removing the step of gaining experience.

[–]aruisdanteSoftware Engineer (14 YOE) 0 points1 point  (0 children)

For sure, I agree completely. This sub optimal “L3 leading L3’s” org at least had the potential to grow within its own control. It’s not so different from an early stage startup founded by a bunch of recent college grads. And some of those turn out to be pretty successful.

On the other hand, with AI agents, it’s not your internal learning loop that improves the output. You can’t mentor them or manage them into better outcomes. They can’t learn from the experience of failure. You’re completely at the mercy of the company that produces the model to put out a better one. And they generally aren’t training it on things that didn’t work. And definitely not on the things that didn’t work for you.

Then again, this state isn’t really new or exclusive to AI. It exist in the human world too. I’m sure anyone that’s worked with off-shored engineering teams in low cost of labor countries has had similar experiences of attempting to manage a team of L3’s who never, never get better at doing their jobs, because anyone with the ability to do that would already have left that low cost of labor country.

Which goes back to the ultimate decision maker: is the cost savings in labor greater than the inefficiency imposed by the quality of that labor?

AI’s cost of labor is so outrageously low that this inflection point has been pushed waaay higher up the food chain than anyone ever would have imagined before.

[–]StorKirken 1 point2 points  (2 children)

You are right, and that is the big challenge we’re facing. How do you develop and maintain both levels of competence?

When we were starting out, the meme was ”rubber duck programming” / explain it to the duck. I think future cohorts of programmers will need to to adapt that or find a new set of practices and snappy phrases to remember them.

[–]kr0n0sShrugg3d[S] 0 points1 point  (1 child)

Why do that when I can spawn an agent team, one to write code and other to be the rubber duck and third one to do PE review? /s

But yeah new practices definitely need to evolve. The current problem with that is the exponential nature of launches and capabilities. Before someone comprehends it the next tech evolves.

[–]StorKirken 1 point2 points  (0 children)

If that would solve the issue and produce a good product, no harm in it, honestly! The trick is evaluating that, though - and unfortunately a lot of lessons are only learned once you hit a brick wall of real consequences. For example, debugging and taking the AI summary at face value: they probably need to experience a few cases where they themselves see how it leads them down the wrong path, and reflect.

I’m also passionate about mentoring interns, but even before AI I’ve seen people get stuck in an analysis and not thinking deeper about it. So I find the current situation very familiar, mostly the same actually! Just on a completely new scale.

Honestly the key question might be the thing you already said: ”why don’t I cut out the middle man and talk to the AI instead of you, intern?” Obviously said in a nicer and more pedagogical way, but something to make them realize and motivate what part they contributed, aside from their prompts.

[–]moh_kohn 3 points4 points  (2 children)

Except engineering managers have not, historically, reviewed code in order to go "why have you written this function three times and why are you parsing html with regex?"

[–]t-tekin 0 points1 point  (1 child)

Depends on the engineering manager and the trust level they have towards you.

[–]moh_kohn 0 points1 point  (0 children)

If engineering managers do do that, then contra the person I was replying to, the kids do need to learn to code.

[–]StTheoSoftware Engineer 1 point2 points  (2 children)

It’s really depressing if that’s the way we’re going. I’m happiest whenever writing code, and a lot of my self esteem came from my work. And that’s assuming that “genuine” AI-backed layoffs never happen…

[–]aruisdanteSoftware Engineer (14 YOE) 1 point2 points  (1 child)

I think there will still always be a place for skilled engineers to write code.

It’s just going to shift to exclusively the kind of problems you stick L7+’s on solving today. The ones that no amount of “well executing junior engineers” can solve. Where writing volume of code doesn’t matter, and instead it’s writing the few lines of correct code.

The problem is that we’re not training anyone to replace the L7+’s of today. Once we’re all retired… it’s a scary thought, to be sure.

And yeah, it really sucks for all the people that had no interest in becoming managers/architects in the first place. It’s an entirely different set of skills. For some, “writing the semicolon” was the actual fun part, they have no interest in turning into the person that “tells the person to write the semicolon”

[–]maxmax4 -2 points-1 points  (0 children)

This is exactly what I’m betting on for my own career. I am focusing on the hard problems and aggressively delegating to AI what is not. Luckily I work in a field where AI is bad at solving the hard problems and where there are many. It’s already a niche field and there’s surely going to be a lack of upcoming talent who are knowledgeable enough to solve those problems.

[–]Vegetable-Score-3915 4 points5 points  (1 child)

I believe you. But I think it is even more important for you to encourage ppl to not rely on AI as their first pass etc working on problems. The ppl that do end up listening to your advice, they will get more out of it.

I've taught coding bootcamps before chatgpt 3.5 came out and after. Students who relied too heavily on ai instead of learning fundamentals, it seemed very similiar to someone just copying and pasting code. They couldn't fix issues despite producing relatively advanced code. It was hard trying to encourage them to be uncomfortable with working through problems with a lot of uncertainty. Ie uncertainty was not considered part of the learning process as much, they were more primed for spoon feeding.

This wasn't a generational thing. Had students from 18 to 55. There was more of a consistent change as students got to start trusting ai tools. My students who were more than half way when chatgpt 3.5 came out, they didn't have the same level of trust and it was treated like stack overflow, it was not as useful as the models we have access to now, but they seemed to really use it as a learning tool and get a lot out of it.

A year later different students, used to these tools from the start, a lot less resilience. Just use to things working straight away and be fixed straight away. Perhaps coding bootcamps probably aren't the best example because of the time pressure and too much content in a short time.

I've also had a vibe coding junior in a previous role. They seemed to place showing competence instead of learning, which is fine in front of clients and the right stakeholders, but their is a time and a place. That was more that person's personality, my point is, it was a lot easier for them to fake it and themselves not be able to tell the difference. I dont know if that is a trend.

I think interns are generally different to new hires, in terms of wanting to want to land a role etc. That is my inference, haven't been a junior for a while, I may be wrong. It is a really tough market.

[–]beefyweefles 4 points5 points  (0 children)

I'm glad you posted this because your experience lines up with what I've seen juniors do at work.

Particularly the part about being afraid to learn in favor of competence. I get that no one wants to show weakness, this has always been the case, but there should be recognition that something quite dangerous is happening. Critical systems (Amazon for one, countless others) are being built on sand and empowering shamelessly inept fakers in ways we haven't seen before.

[–]Special_Rice9539 4 points5 points  (0 children)

It took me many years to understand how critical this part of swe is and why it makes swe so powerful.

A lot of jobs are just about shipping things out, but in swe, your product might bring in a couple hundred million dollars a year, and it’s more valuable for you to spend all year exploring and make one multi-million dollar change than hustle and grind making 10k wins.

It kind of allows for a beautiful mentor-apprentice relationship that is lost elsewhere as you focus on mastering your craft.

I think interns do want to learn, but they’re intimidated by failure, so it’s hard to resist having AI do everything. I think really emphasizing to them that you’d rather have them struggle and fail so they learn than ship something it’s important.

Let’s be real, their intern project is likely not going to be used anyways, or it’s something you could do in a day, so they need to know what the real point of the project is

[–]dronz3r 7 points8 points  (4 children)

Unfortunately, the way we saw software engineering skills evolve with experience has completely changed now. It takes years of experience to learn to write good code, all this is mostly obsolete now. Everything that a junior engineer is supposed to do is being done by AI. I also would rather speak to AI to generate the code than have juniors prompting and committing the same code.

It's frustrating but there isn't any way around it unfortunately.

[–]mainframe_maisie 7 points8 points  (1 child)

I don’t think it really is obsolete. You still need experience and intuition to be able to notice code smells and issues and bugs that AI generated code introduces, and it’s more a case of “when” than “if”. And the cost/benefit calculations are going to need to be redone once AI companies increase their prices by 10x in the next year.

[–]-manabreak 8 points9 points  (0 children)

Very true. In just a year, some of the most senior developers at work are turning into AI supervisors without knowledge of the code. Just last week, they were wondering why our memory usage has increased by 50% between releases, and they couldn't say a single thing. These same guys were micro-optimising the codebase a year ago.

[–]kr0n0sShrugg3d[S] 5 points6 points  (1 child)

Yes and as the Ai Models and Agentic harness on top of it evolves, best practices might be auto enforced.

After replacing all other jobs with automation, for their last act of hubris, the developers replaced themselves.

[–]dronz3r 1 point2 points  (0 children)

Indeed, developers need to think more like product and business owners now.

It may be a good thing, barrier to start software firms reduces and eventually we'll have more competition reducing the cost of software.

[–]hoxxii 1 point2 points  (0 children)

At my first job my boss didn't even say hi (he sat infront of me) and peers got their coffee behind a closed door which I only learned a month in cause I didn't understand why I sometimes sat there alone.

All of you who love to mentor are angels in my eyes and I hope this hype cycle doesn't kill that off completely. You all are needed.

[–]Marceltellaamo 1 point2 points  (1 child)

Senior dev here, I have seen the same. The issue isn’t AI, it’s interns outsourcing thinking too early. If they don’t struggle through first principles at least once, nothing sticks. Curious how people are enforcing "no AI on first pass" without killing momentum.

[–]Wonderful-Habit-139 2 points3 points  (0 children)

There's no momentum if the AI generated code has to go through a PR review.

[–]No_Weight4426 2 points3 points  (4 children)

If they are like this, how come the pass the interview process? Wouldn't they have been flagged during the interview?

[–]kr0n0sShrugg3d[S] 4 points5 points  (3 children)

There are usually quotas for interns based on colleges. And the hiring happens with DSA which is very different from production grade code, internship gives that realignment.

Plus the interns are smart, Ai just exacerbates the curse of the gifted programmer.

[–]budding_gardener_1Senior Software Engineer | 12 YoE 8 points9 points  (1 child)

And the hiring happens with DSA which is very different from production grade code, internship gives that realignment. 

can you please explain that to the people who conduct fucking leetcode interviews

[–]kr0n0sShrugg3d[S] 3 points4 points  (0 children)

Beats me man, I fucking hate them. I usually start with a very simple problem and then add complexities and tradeoffs we see in a job. But then there are folks asking travelling salesman problems and LC hard questions.

[–]lightreee 1 point2 points  (0 children)

Is this AI push the same idea as Gen Z not ‘understanding’ computing as they are mobile-first?

I was born in 1993 and did a lot of deep hacking and programming since I was like 12. There wasn’t any tools or help so i had to learn

Seems like the changes to tooling making it much easier has allowed a wider set of knowledge levels for people to be part of SWE

[–]Outside-Storage-1523 1 point2 points  (0 children)

I think the entire purpose of intern is to learn professional coding and looks like they are exactly doing that /s.

OP please blame your leaders who force people to use AI at work.

[–]LeopattoCEO / Data Scientist, 8+ YoE 0 points1 point  (0 children)

As I said before and ill say it again - people don't want to be taught, they want to do their 8 hours and fuck off home to drink beers.

If the AI codes the feature for them so be it, it's not your job to teach; your job as a senior is to mostly review the code and fix any critical mistakes.

[–]a_protsyuk 0 points1 point  (0 children)

The tell is whether they can explain the wrong answer. If their response to a bug is opening a chatbot rather than forming a hypothesis - even a wrong one - then there's no debugging muscle forming. Tool proficiency without a mental model of how systems fail is just copying with extra steps.

[–]tictacotictaco 0 points1 point  (0 children)

It drives me fuxking crazy when I’m reviewing a JR’s code, and they can’t explain what a certain part is doing. It’s so much harder reviewing JR AI slop.

We recently interviewed a new gen coder. Uses AI almost (maybe totally) exclusively. Scary stuff.

[–]Irish_and_idioticSoftware Engineer 0 points1 point  (1 child)

Can you help me understand what a tier 0 service is? Are we talking bare metal? VMs sorry I know that’s not the point of post but I have never heard that term before.

I’ll read rest of post

[–]kr0n0sShrugg3d[S] 0 points1 point  (0 children)

Tier 0 services are the bedrock on which the other services are built on and other services are dependent on. These services are statically stable and have the least amount of external dependencies. These can be your compute, blob storage, networking.

[–]qwerty8082 0 points1 point  (0 children)

Damn you sound like the best ✨

[–]DigmonsDrill 0 points1 point  (0 children)

Third time is the charm posting this.

[–]arihoenig 0 points1 point  (0 children)

If they have zero intellectual curiosity, why did your company offer them an internship?

[–]Decent_Muffin_7062 -1 points0 points  (7 children)

Sorry to be blunt but your interns sound incompetent - how are you hiring these people at the supposed 'biggest cloud provider'? I'm a large non-tech company, and while we have a few who do this they are swiftly disciplined/don't get a return offer.
Even before AI, there have always been people who produce slop with no knowledge. Boilerplate copy pasted from elsewhere, random crap that doesn't even work they just want to submit a PR.
They've always had to explain themselves and in places with low standards nobody bothered until a production outage occurred. Same thing,

The other big thing (at my company at least) is that you're responsible for AI output. So if you don't read it, and merge a PR that causes issues, you're held responsible. Doesn't your company have an AI use in SWE policy?

[–]kr0n0sShrugg3d[S] 6 points7 points  (6 children)

The interns are smart kids, they are feeling the pressure to deliver and like any smart person they are taking the path of least resistance. I am seeing this across the board. That is what makes it more frustrating.

Every company is going crazy after the output volume right now. There was a recent outage that was caused due to Ai written code, so there might be a change in policy.

[–]bcameron1231Principal Engineer 6 points7 points  (4 children)

If they are feeling pressure to deliver (regardless of the environment outside your walls), it sounds like your organization needs to establish what an internship means, and create an environment where interns learn it's about growth and knowledge, not output. A culture where failure isnt just an option, it's expected as a teaching mechanism.

You pay full-time employees for output, not your interns. Your work environment should reflect that.

Establish baselines and rules. Advocate for AI as a tool, but not a replacement. Periodically do code reviews together, If they can't explain the code, it doesn't get merged. Set up pair programming sessions to work along side them. Give them tasks, but before doing hands on keyboard, give them time to research and explain an approach to you so you can discuss and work together... Gives you a chance to challenge them, play devil's advocate, and have real thoughtful discussions...even if AI ends up doing much of the work later.

Good internship programs take time and effort. Seems like you're giving in to early. If mentorships is what previously drove you, find ways to build a growth culture. You can do it, I believe it.

[–]Wonderful-Habit-139 2 points3 points  (1 child)

Agreed. When I was an intern, my mentor actually made me slow down to allow me to get used to the environment and take more time learning and growing.

[–]bcameron1231Principal Engineer 1 point2 points  (0 children)

💯 the way to go about it, imo.

As an intern, you don't know what you don't know. You put a huge amount of pressure on yourself and think we're expecting you to just start and deliver work. We don't even expect that from Senior new hires! So we need to establish that early.

I've been in the mentorship space for a long time, I always try to reiterate that, I hire people for their traits. I want people (interns or Seniors) to have a growth mindset, a want to learn from failure, and someone I can enjoy working with. 9/10 I'll choose that person versus the know-it-all or the code crusader. I'm a firm believer that value doesn't only come from productivity.

Internships don't have to lose their value because AI is here. It's just another tool in the box. We can still create an environment for learning.

[–]kr0n0sShrugg3d[S] 1 point2 points  (1 child)

Interns are told that they would get a return offer if they deliver, earlier this used to be a driving motivation to learn, but with Ai this has led to interns shortcircuiting the entire process. Feels like we need to realign this thought process.

When metrics become the goal ...

And you are right, will not give up on it but evolve the process.

[–]bcameron1231Principal Engineer 1 point2 points  (0 children)

I can't disagree with you.

All I was trying to say that's a company culture problem, not an AI problem.

If you're passionate about it, you can do it. Get buy in from your colleagues. You want to offer jobs to people you like, are competent, and willing to work hard. If you establish that culture, I promise you, you'll reap the rewards when these interns become your full time colleagues.

Keep your head up. Young folks entering Software Engineering, need mentors and people like yourself, more than ever.

[–]Decent_Muffin_7062 1 point2 points  (0 children)

Let me put it another way. There are plenty of smart people who aren't curious and don't even like coding. These people will just use AI and go home because they want the most money with the least effort. There's so much hype about just using AI to build an app and become millionaires, they believe it.
Others are curious - true builders, and will want to understand how things work because they like the craft.

Maybe I'm just oblivious, I've never worked for Big Tech. But I find it hard to believe that your company puts so much pressure on interns to deliver a certain volume, that they don't have time to even read and understand their code. Or that your opinions don't impact the return offer. Again where I work they are measured on behaviours not just output, and the opinions of seniors they worked with. Interns are there to learn and not measured on the same things as a permanent employee.

[–]titan_bullet -3 points-2 points  (3 children)

I would see this as a good thing rather than the "death of a passion". Be up-front with them - no AI, you need to do this on your own, and then explain your reasoning.

If they ignore that and you detect that they use AI (which should be pretty clear from a quick chat with them on a pair PR review), you can use that as a filtering mechanism for which trainees are worth your time more.

[–]Calamero -1 points0 points  (2 children)

It’s like letting the apprentice not use power tools or cnc machines but having them carve wood by hand.

Cool if your company can afford that. Not viable for most businesses.

[–]bcameron1231Principal Engineer 1 point2 points  (0 children)

You don't hire interns for their output and productivity.

You hire interns as a future pipeline of prospective employees that you build up, can trust and want to work with. In almost all cases, internships are a net loss in productivity and money.

If you're hiring interns for their output, that says a lot more about the companies business practices and seeking cheap labor. If they can't afford it, they shouldn't be hiring interns, imo.

[–]greensodacan 0 points1 point  (0 children)

Just pointing out, that's exactly what journeymen have apprentices do. It's important that they understand how wood fibers behave, how grain works, how wood contracts and expands, where/when/how to drive a nail, etc.

I have a lot of carpenters in my family and they still break out files and chisels every now and then. Sometimes someone screws up a cut, or it's an unusually warm day and the wood isn't behaving, or some corner isn't actually square, the list goes on.

[–]JobStrange153 -1 points0 points  (0 children)

focus on developing unique problem-solving skills

[–]Nofanta -4 points-3 points  (0 children)

Good thing. Passion should be reserved for your personal life.